Anti-Aliasing Performance
While we didn't get chance to test all of the new supersampled anti-aliasing modes, we did test the custom filter AA modes in rather comprehensive fashion. The custom filters haven't changed in the transition from RV770 to Cypress/RV870, so image quality is effectively unchanged.
Effectively though, the Radeon HD 5870 has some anti-aliasing modes that are useful, and others that actually have a detrimental effect on image quality. We recommend not using the Narrow Tent or Wide Tent (NTAA and WTAA) filters, as they soften edges in the scene and make images appear a little blurred (more so at higher AA levels).
Both NTAA and WTAA are very crude ways of achieving higher levels of anti-aliasing without killing performance, and we'd take AMD to task on this if it didn't also offer alternative high-quality anti-aliasing modes. We recommend sticking with the regular box filters and, if you've got enough performance on tap in the game you're playing, it's worth giving the edge detect filter a try, as this is the one custom filter AA mode that doesn't soften edges. The edge detect filter is nothing new, but it's worth pointing out again for the sake of completeness.
See page 14 for test kit
The Radeon HD 5870 bumps into
Fallout 3's frame rate cap at the lower end of the anti-aliasing scale, but it really shows its horsepower at the higher end, as even 8x Edge-Detect AA delivers an average frame rate of 57.1fps at 1,920 x 1,200. MSAA performance is strong, too, although it's difficult to deduce whether Cypress's MSAA resolve units are more efficient than RV770's.
The penalty for enabling 8x AA on the Radeon HD 4890 was hardly massive (it's just a 10fps hit on the average frame rate, although minimum frame rates do drop by nearly 50 per cent) and we expect the Radeon HD 5870 to be similar in that respect. We tried to get the various custom-filter AA modes working properly in
Crysis as a more intensive test, but our results ended up being very inconsistent and so the test had to be dropped.
Want to comment? Please log in.